Summary: Unhealthy alcohol consumption is embedded within people?s everyday lives, but it is difficult to study individuals outside of laboratories and treatment offices. Many individuals engaging in excessive alcohol consumption do not make it to treatment until it has had large, sometimes catastrophic, negative effects on their life. Mobile phone apps and social media, with care taken for consent and privacy, offer an avenue for large-scale behavior-based study within an ecological context. This proposal seeks to develop techniques for the study of and prediction of unhealthy alcohol consumption within the real-world context of the restaurant industry, a population where excessive alcohol consumption is highly prevalent. Using innovative and rigorous data science techniques, we will study the cross-sectional, prospective longitudinal, and community-based relationships between unhealthy drinking and (a) affective states, (b) stress, and (c) two types of empathy: depleting and beneficial. In the process we will: (1) build a large and secure registry of digital mobile data (N = 5,925) about drinking behavior, (2) evaluate existing data-driven assessments of psychological states, (3) use machine learning to improve assessments of psychological states and predict future drinking behavior, and (4) perform one of the largest scale studies, to date, of the relationship between psychological state and unhealthy drinking. Our specific aims include: (1) Automatically assess the association of unhealthy alcohol consumption with affect, stress, and empathy among restaurant industry workers based on their linguistic behavior in social media and text messaging; (2) Develop a mobile app for longitudinal collection of fine-grained daily psychological health to analyze relation to and build prospective predictive models of daily drinking patterns; (3) Examine community affect, stress, empathy, and open-vocabulary factors, as represented by millions of local posts on public social media and assess their relationship to individual drinking behavior for restaurant industry workers. Each aim includes both the development of computational research tools and the testing of specific hypotheses. Constructs range from those with an extensive literature with respect to unhealthy drinking (emotional states), to those with burgeoning and conflicting research (stress), to those that are highly novel (empathy). We have extensive experience in collecting data and developing apps, including preliminary work at recruiting bartenders and servers. Related research and our preliminary work already suggests that there are strong links between unhealthy drinking and digital language data. We will release our software tools -- the app platform and predictive models -- under open source licenses accompanied with instructional tutorials. We see this work as trail-blazing a broad use-case for data scientific language-based assessments to study unhealthy drinking.